Backend systems reportedly fabricated fake "likes," profile visitors and pre-recorded video calls whenever no real gig worker was available.
A China-based app development studio used Anthropic's Claude models to run a network of more than 20 fraudulent dating apps, deploying thousands of AI-generated personas to converse with real users while advertising the platforms as staffed entirely by real people, Anthropic's latest threat intelligence report has revealed.
Over a two-week window in April 2026, Anthropic, the American artificial intelligence research and safety company behind the Claude family of large language models, said it identified more than 4,700 distinct AI personas that held conversations with at least 25,000 unique individuals across the app network, generating roughly 2.36 million messages in that period alone.
According to the report, the operation blended a small number of real people, recruited as gig workers, into the same matching feed as the AI-generated profiles, at a ratio of roughly three AI personas for every one human.
The human workers were used specifically to defeat scepticism, handling tasks the AI could not, such as live video calls and social media follow-backs, while a separate, smaller AI model even drafted reply suggestions for the human workers to send.
